108 OLMo-style LMs (4M-75M) with ~3.9K checkpoints from ICML paper released on Hugging Face
BrihiJ · x · 2026-10-15
The dill-lab team has publicly released the 108 OLMo-style language models (4M–75M) from their ICML 2026 paper on Hugging Face, in the GeoLadder collection. The models were trained across controlled sweeps of tokens, batch size, learning rate, weight decay, and LR decay, and ship with roughly 3.9K intermediate checkpoints — a full controlled dataset for studying training dynamics and scaling laws.
Related event: dill-lab Open-Sources 108 Small Language Models with ~3,900 Checkpoints(2 posts)→
More from Research
- OpenAI's Quasi-Riemann proof sparks chain of results, Bhargava proves primitive root theorem — Singularitarian · 2026-10-15
- Applied Compute's Spork post-training recipe doubles chess puzzle solve-rate vs GRPO — ypatil125 · 2026-10-15
- AAAI's AI Magazine Fall 2026 issue (Vol. 47, No. 3) released with 20+ articles — FrnkNlsn · 2026-10-15
- Geometric self-attention: 14,000x smaller state than KV cache with better CE at 2K context — iskander · 2026-10-15
- Stanford's CausalSmith automates causal inference research with Lean-verified proofs — StanfordAILab · 2026-10-15
- Cellular Intelligence wins Best Poster at scverse 2026 for perturbation platform — arjunrajlab · 2026-10-15